Venue : Centre Broca
Thesis defended in english
Milesa Simic
Team : Network dynamics for procedural learning
IMN
Thesis directed by Marc Deffains
Title
Distinct dorsal and ventral basal ganglia contributions to value-based decision-making in health and parkinsonism
Abstract
Value-based decision-making (DM) relies on processes that evaluate options by assigning them a subjective value, guide action selection and execution, and evaluate consequences by comparing expected and obtained outcomes to update decisions. These processes rely on coordinated activity within distributed brain networks including the basal ganglia (BG). The BG comprise interconnected nuclei modulated by midbrain dopamine and organized into partially overlapping cortico-subcortical circuits along a dorsal-ventral axis. Dorsal circuits are associated with motor and cognitive control, while ventral circuits are linked to valuation/evaluation and motivation. Dysfunctions of these circuits have been implicated in DM deficits observed in Parkinson’s disease (PD). However, their distinct contribution to value-based DM in healthy and pathological conditions remains unclear.
To address this question, we performed electrophysiological recordings before and after PD induction in key BG nuclei (caudate nucleus, nucleus accumbens (NAc), external globus pallidus and ventral pallidum) in non-human primates performing a value-based DM task in which expected values (EVs) varied with reward magnitude and probability. We examined how decision variables (value- and action-related signals) associated with losses and gains are encoded and organized across BG structures. Behavior scaled with EV: higher EV options were chosen more accurately, elicited fewer omissions and were associated with faster decision times, linking valuation to behavioral engagement and vigor. Single-unit recordings revealed a functional gradient across structures: dorsal regions preferentially encoded action-related parameters, while value encoding increased ventrally. EV representations were non-linear with maximal sensitivity at the loss-gain transition. Population-level neural dynamics revealed distinct neural trajectory geometries across EVs, particularly at this transition in ventral regions, where trajectory separation and divergence increased, while spatial signals were more segregated dorsally. Decoding analyses supported this dissociation: EV was best decoded from ventral populations, whereas spatial parameters were most accurately decoded from dorsal populations. Thus, representation of value and action signals across BG structures is distributed but organized.
To examine DM deficits in PD, we induced progressive dopaminergic depletion using chronic low-dose 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) and compared healthy, early, and late disease stages. Dopamine loss led to marked task disengagement and reduced movement vigor, accompanied by changes in neuronal encoding despite preserved decision performance. The most pronounced alterations occurred in the ventral striatum, including a late-stage decrease in value encoding in NAc neurons, while pallidal representations of decision variables were largely intact. Analysis of cue-evoked changes in neural response magnitude (neural gain) showed maintained differentiation between performed and omitted trials and revealed systematic co-shifts between disengagement and the increasing prevalence of omission-like low-gain neural states across disease stages. Dopamine depletion may therefore shift global network states (from engaged to disengaged) rather than abolish valuation.
Together, this work supports a functional organization of the BG during value-based DM in which ventral circuits preferentially encode value, including non-linear loss-gain sensitivity, while dorsal circuits preferentially encode action-related variables. Dopaminergic depletion primarily affects motivation and task engagement through shifts in global neural states and selective vulnerability of ventral striatal circuits, while pallidal populations may help sustain accurate decision performance by maintaining representations of task-relevant decision variables.
To address this question, we performed electrophysiological recordings before and after PD induction in key BG nuclei (caudate nucleus, nucleus accumbens (NAc), external globus pallidus and ventral pallidum) in non-human primates performing a value-based DM task in which expected values (EVs) varied with reward magnitude and probability. We examined how decision variables (value- and action-related signals) associated with losses and gains are encoded and organized across BG structures. Behavior scaled with EV: higher EV options were chosen more accurately, elicited fewer omissions and were associated with faster decision times, linking valuation to behavioral engagement and vigor. Single-unit recordings revealed a functional gradient across structures: dorsal regions preferentially encoded action-related parameters, while value encoding increased ventrally. EV representations were non-linear with maximal sensitivity at the loss-gain transition. Population-level neural dynamics revealed distinct neural trajectory geometries across EVs, particularly at this transition in ventral regions, where trajectory separation and divergence increased, while spatial signals were more segregated dorsally. Decoding analyses supported this dissociation: EV was best decoded from ventral populations, whereas spatial parameters were most accurately decoded from dorsal populations. Thus, representation of value and action signals across BG structures is distributed but organized.
To examine DM deficits in PD, we induced progressive dopaminergic depletion using chronic low-dose 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) and compared healthy, early, and late disease stages. Dopamine loss led to marked task disengagement and reduced movement vigor, accompanied by changes in neuronal encoding despite preserved decision performance. The most pronounced alterations occurred in the ventral striatum, including a late-stage decrease in value encoding in NAc neurons, while pallidal representations of decision variables were largely intact. Analysis of cue-evoked changes in neural response magnitude (neural gain) showed maintained differentiation between performed and omitted trials and revealed systematic co-shifts between disengagement and the increasing prevalence of omission-like low-gain neural states across disease stages. Dopamine depletion may therefore shift global network states (from engaged to disengaged) rather than abolish valuation.
Together, this work supports a functional organization of the BG during value-based DM in which ventral circuits preferentially encode value, including non-linear loss-gain sensitivity, while dorsal circuits preferentially encode action-related variables. Dopaminergic depletion primarily affects motivation and task engagement through shifts in global neural states and selective vulnerability of ventral striatal circuits, while pallidal populations may help sustain accurate decision performance by maintaining representations of task-relevant decision variables.
Key words
Basal ganglia, decision-making, value, electrophysiology, non-human primate, Parkinson’s disease
Jury
Mme Yulia WORBE, Professeure des universités – praticienne hospitalière, Institut du Cerveau (Paris), Rapporteure
M. David THURA, Chargé de recherche, Centre de Recherche en Neurosciences de Lyon (Bron), Rapporteur
Mme Shauna PARKES, Directrice de recherche, Institut de Neurosciences Cognitives et Intégratives d’Aquitaine (Bordeaux), Examinatrice
M. Jérôme MUNUERA, Chargé de recherche, Institut du Cerveau (Paris), Examinateur
M. David THURA, Chargé de recherche, Centre de Recherche en Neurosciences de Lyon (Bron), Rapporteur
Mme Shauna PARKES, Directrice de recherche, Institut de Neurosciences Cognitives et Intégratives d’Aquitaine (Bordeaux), Examinatrice
M. Jérôme MUNUERA, Chargé de recherche, Institut du Cerveau (Paris), Examinateur